*2.2. Data Processing and Classification Method*

This study uses satellite images and auxiliary data, and employs a hybrid classification approach for image classification. The auxiliary data includes those from field investigation, yearbooks, local chronicles, Google images, and the land use/land cover maps of the Resource and Environment Science and Data Center (https://www.resdc.cn (accessed on 13 December 2021)). The auxiliary data were mainly used to identify training samples for image classification, testing samples for accuracy evaluation, and visually modified classified images. A hybrid classification method, which was developed for image classification, was found to be an effective approach to enhancing the accuracy of image classification [34]. This method includes automatic classification (random forest) and visual modification. The classification process is shown in Figure 2. See the following steps for details.

#### 2.2.1. Data Processing and Classification Method

This study adopts the physical boundary of islands as the division, including the ecological and utilized areas of the islands' intertidal zones and neritic regions. The Geospatial Data Cloud (http://www.gscloud.cn/ (accessed on 15 December 2021)) server was used to download Landsat 8 Operational Land Imager (OLI), and Thermal Infrared Sensor (TIRS) database was used for the images of the 2014s and 2015s. All images used in this study are from the dry season period (November–January), during which zero cloud cover allowed for high image quality. The raw images have a spatial resolution of 30 m.

Preprocessing is an essential step to correct atmospheric effects and minimize geometric and radiometric errors before image classification. This study used the ENVI software to undertake the radiometric calibration and atmospheric correction of all the images. A fusion or pan-sharpened multispectral (MS) image provides an improved image of high

spatial resolution and can also improve the classification results [35,36]. Therefore, the panchromatic (PAN) band with a resolution of 15 m and MS images were adopted to be fused by the Gram–Schmidt (GS) method in this study. Subsequently, four pan-sharpened MS images with a spatial resolution of 15 m of the studied area were generated.

**Figure 2.** Flow chart of island land-use classification in the study area.
